Statistical Data Format Converter

Problem

Statistical and clinical-trial data are often locked across SPSS, Stata, SAS, Excel, Parquet, and similar files. A plain pandas read may preserve values but drop variable labels, value labels, special-missing-value meanings, and other metadata. Cross-format conversion makes this harder when tools disagree about which metadata can survive.

How It Works

The skill separates extraction from conversion. It first reads a source file into a pandas-compatible structure and reports which metadata is retained or lost. It then converts among common statistical formats or exports to Parquet, HDF5, JSON, CSV, Excel, and related targets. For binary statistical formats, it aims to preserve variable and value labels where supported. For Parquet and Arrow, labels can be embedded in schema.metadata; for CSV/TSV, a sidecar _metadata.json file helps keep metadata alongside the data. The default behavior is preview-and-warn: R-backed paths are opt-in and disabled unless allow_r_exec=True is explicitly set.

Boundaries

This is not a general-purpose ETL engine, and not every supported format is lossless. Some formats are detect-only, require optional R backends, or preserve only partial structure. Output should still be validated, especially before regulatory submission.